> ## Documentation Index
> Fetch the complete documentation index at: https://docs.bytespike.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# DOSIA
> DOSIA is ByteSpike's first-party desktop AI client. Two modes: personal (you bring a ByteSpike key) or enterprise (admin-provisioned, lark SSO, auto-allowed models).
DOSIA is ByteSpike's own desktop client — a multi-model chat app with
image/video generation, vision, MCP tool calling, and an agent loop.
It speaks every protocol ByteSpike serves, so the same key reaches
Claude, GPT-5, Gemini, DeepSeek, Doubao, Kimi, GLM, MiniMax, Nano
Banana, Veo — without per-vendor reconfiguration.
Two ways to wire it up:
* **Personal mode** — install DOSIA, paste your own `sk-byts-...` key, pick models. This page covers that path.
* **Enterprise mode** — admin pre-provisions you via lark SSO, DOSIA auto-picks up the allowed model set, no manual key. See [Employee onboarding](/employee-onboarding) for the enterprise flow.
## Install
Download the latest dmg from
[github.com/leave1206/dosia-releases](https://github.com/leave1206/dosia-releases/releases/latest):
* Apple Silicon (M1/M2/M3/M4): `DOSIA-x.y.z-arm64.dmg`
* Intel: `DOSIA-x.y.z.dmg`
Drag DOSIA into `/Applications`, then launch.
First-launch may report **"file is damaged"** or **"from an
unidentified developer"** — that's macOS Gatekeeper blocking an
un-notarized dmg, not a real issue. Fix:
```bash theme={null}
xattr -cr /Applications/DOSIA.app
```
Then relaunch.
Builds tracked in the same release page. If your platform isn't
listed, file a request in the repo's issues — Mac is the primary
target today.
## Personal mode setup
[console.bytespike.ai/keys](https://console.bytespike.ai/keys) → **Create key**.
Pick the routing group based on what you want DOSIA to reach (`claude-default` for Claude, `gemini-default` for Gemini, or `default` to route everything).
See [Models](/models) for the group-vs-model map.
Top menu: **DOSIA → Settings** (shortcut `⌘,` on Mac).
**Deploy Mode** stays on **Personal** (the default). Enterprise mode is for SSO-managed accounts — see [Employee onboarding](/employee-onboarding) if your admin gave you a backend URL.
**Settings → AI Models → Add Provider → "ByteSpike"**.
| Field | Value |
| -------- | ----------------------------------------------------------------------------------- |
| Base URL | `https://llm.bytespike.ai` |
| API key | your `sk-byts-...` |
| Protocol | Anthropic Messages (or OpenAI Chat Completions — either works against the same key) |
DOSIA will hit `/v1/models` against ByteSpike to fetch the
list of models your key can actually reach. Wait a couple seconds
for the dropdown to populate.
Back in the main chat box, pick any model in the dropdown ("Claude Sonnet 4.6" / "GPT-5.5" / "Gemini 3.5 Flash" / etc.). Type a message, hit enter.
## What works in DOSIA
| Feature | Backed by |
| -------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| Text chat | All text models (Claude / GPT / Gemini / DeepSeek / Kimi / GLM / MiniMax / Doubao) |
| Vision (drag image into chat) | DOSIA's `analyze_image` MCP tool calls a vision model (Claude / GPT-5-4 / Gemini) |
| Image generation ("draw a ...") | `generate_image` MCP tool routes to Nano Banana / GPT Image |
| Video generation ("generate a video of ...") | `generate_video` MCP tool routes to Veo |
| Agent (multi-step task) | DOSIA's agent loop calls `tool_use` / function calling across providers |
| MCP plugins | DOSIA bundles 9 official plugins (Lark, browser automation, meeting analysis, knowledge base, etc.) — see [github.com/leave1206/dosia-plugins](https://github.com/leave1206/dosia-plugins) |
For the routing model (how the agent decides which model handles
which step), see [DOSIA Agent mode](/concepts/dosia-agent-mode).
For the MCP bridge between DOSIA and the model side, see
[MCP integration](/concepts/mcp-integration).
## Switching models
Each chat thread remembers the model it was opened with. Switch at
the top of the chat box — the dropdown shows every model your key
can reach. Image / video model picks happen inside the respective
tool calls; you don't need to set them per-thread.
## Troubleshooting
| Symptom | Cause | Fix |
| ---------------------------------------------- | ------------------------------------------------- | ------------------------------------------------------------------------------ |
| "Failed to fetch models" after adding provider | Key invalid or wrong group | Copy a fresh key from console; confirm group serves the models you expected |
| Model dropdown empty | Key has 0 allowed models | Check the key's `group_id` in console; pick a non-empty group |
| Image / video tool button greyed out | The MCP plugin isn't enabled in DOSIA settings | **Settings → MCP** → enable `image-tools` / `video-tools` |
| Vision drag-drop returns 400 | The chosen text model doesn't support image input | Switch the chat to `claude-sonnet-4-6`, `gpt-5-4`, or `gemini-3-5-flash` |
| "insufficient\_balance" | Key (or org wallet) is out of credits | Top up at [console.bytespike.ai/billing](https://console.bytespike.ai/billing) |
| Slow first token | Cold start on a cross-protocol translation route | Warm with a small request first, or stick to a native protocol for hot path |
## Enterprise mode (lark SSO)
If your company runs DOSIA centrally: admin provisions the enterprise
backend URL, DOSIA flips into Enterprise mode, lark SSO logs you in,
the allowed model set syncs automatically — no per-user key. The
full setup is in [Employee onboarding](/employee-onboarding).
The admin side (granting employees, monitoring usage, configuring
allowed groups) is documented in [Admin operations](/admin-operations).
## Next
Enterprise / SSO path.
How DOSIA's agent loop picks models per step.
The bridge between DOSIA's tool plugins and the model side.
Wire any other client (Claude Code / Codex / Cursor / SDK).